Search

The Hengill geothermal field, located in southwest Iceland, is host to the Hellisheiði power plant, with its 40+ production wells and 17 reinjection wells. Located in a tectonically active area, the field experiences both natural and induced seismicity linked to the power plant operations. To better manage the risk posed by this seismicity, the development of robust and informative forecasting models is paramount. In this study, we compare the forecasting performance of a model developed for fluid‐induced seismicity (the Seismogenic Index model) and a class of well‐established statistical models (Epidemic‐Type Aftershock Sequence). The pseudo‐prospective experiment is set up with 14 months of initial calibration and daily forecasts for a year. In the timeframe of this experiment, a dense broadband network was in place in Hengill, allowing us to rely on a high quality relocated seismic catalog. The seismicity in the geothermal field is characterized by four main clusters, associated with the two reinjection areas, one production area, and an area with surface geothermal manifestations but where no operations are taking place. We show that the models are generally well suited to forecast induced seismicity, despite some limitations, and that a hybrid ETAS model accounting for fluid forcing has some potential in complex regions with natural and fluid‐induced seismicity. In the southwest of Iceland, the Hengill volcanic region is the seat of a geothermal field exploited with two power plants. The power plants provide district heating and electricity to the capital region. The area continuously experiences small to moderate earthquakes, associated to the volcanic nature of the region and to the power plant operations (injection and withdrawal of fluids from the underground). To better manage the risk posed by these earthquakes, we use statistics‐based computer simulations to forecast the rate of earthquakes during a year‐long experiment. The simulations are trained on 14 months of data. One of the models (the Seismogenic Index) is designed to relate rate of earthquakes to the volumes of fluids injected and withdrawn; while the other relies on statistical characteristics of earthquake sequences. We show that these computer simulations are well suited to forecast earthquake rates in the Hengill geothermal field, even though they have their respective limitations. The combination of the statistical seismicity simulations with a term accounting for the volumes does show promising results in an area with complex earthquake sequences. Statistical models are used in a pseudo‐prospective experiment to forecast daily seismicity rates in the Hengill geothermal fieldEpidemic‐Type Aftershock Sequence (ETAS)‐type models are able to adapt quickly to sudden changes in the seismicity rateThe Seismogenic Index‐type model performs at least as well as ETAS‐type models outside periods of high‐intensity seismic activity Statistical models are used in a pseudo‐prospective experiment to forecast daily seismicity rates in the Hengill geothermal field Epidemic‐Type Aftershock Sequence (ETAS)‐type models are able to adapt quickly to sudden changes in the seismicity rate The Seismogenic Index‐type model performs at least as well as ETAS‐type models outside periods of high‐intensity seismic activity

We analyze two high‐quality Southern Californian earthquake catalogues, one with focal mechanisms, to statistically model and test for dependencies of the earthquake‐size distribution, the bvalues, on both faulting style and depth. In our null hypothesis, bis assumed constant. We then develop and calibrate one model based only on faulting style, another based only on depth dependence and two models that assume a simultaneous dependence on both parameters. We develop a new maximum‐likelihood estimator corrected for the degrees of freedom to assess models' performances. Our results show that all models significantly reject the null hypothesis. The best performing is the one that simultaneously takes account of depth and faulting style. Our results suggest that differential stress variations in the Earth's crust systematically influence bvalues and that this variability should be considered for contemporary seismic hazard studies. We model the dependence of the bvalue on depth and tectonic style in Southern CaliforniaStatistical tests reveal that faulting style affects the bvalue more than depthSimultaneous dependencies result with high statistical significance in the best data description

The increase of temperature of a low‐permeability, fluid‐saturated media may trigger significant thermal pressurization, where the expansion of pore fluid cannot be accommodated by the thermal expansion of the pore space. With the aid of a coupled thermohydromechanical numerical simulator, we investigate the possible impact of thermal pressurization during the life of a deep geological repository (DGR) for high‐level radioactive waste, characterized by the emplacement of radioactive material‐filled canisters in a series of parallel tunnels, excavated in a low‐permeability clay formation. We represent the fault as a planar structure embedded in a thermoporoelastic material and shear activation evaluated by a strain‐softening Mohr‐Coulomb failure criterion. The results show that stress changes caused by temperature and thermal pressurization of a rock mass around the emplacement tunnels may trigger a slip event on a fault plane in proximity of the geological disposal site: Rupture nucleates at depth, hundreds of meters below the DGR. Stress transfer plays a key role, while a direct hydraulic connection between the repository and the nucleation zone is not necessary in order to trigger rupture. A low stress ratio may favor the occurrence of slip up to a distance of 600 m of the fault from the outermost tunnel. These results highlight the need of investigating hydromechanical properties and local stress conditions at depth to characterize the geomechanical response of weak planes located in the surroundings of a high‐level radioactive waste repository and to provide sufficient knowledge for the safe development of the DGR site. Temperature increase in a low‐permeability formation will lead to pore pressure increaseStress transfer and poroelastic effect can affect negatively the stability of nearby fault sections without direct hydraulic connectionFault reactivation may happen in the first 1,000 years after canisters emplacement

We analyze slip distribution and rupture kinematics of a Mw3.3 induced event that occurred in the St. Gallen geothermal reservoir (NE Switzerland) in 2013. We carry out a two‐step procedure: (1) path effects are deconvolved from the seismograms using an empirical Green's function, resulting in relative source time functions at all seismic stations; (2) the relative source time functions are back‐projected to the corresponding isochrones on the fault plane. Results reveal that the mainshock rupture propagates toward NNE from the hypocenter with an average velocity of 2,000 m/s. Spatiotemporal organization of foreshocks and aftershocks shows that the mainshock broke a previously less active portion of the fault and suggests that the aftershock sequence could be mainly driven by stress transfer. Applying this method in an operational environment could enable fast retrieval of seismic slip, allowing assessment of fault asperities and structures involved in the reservoir creation process. In most models and analyses, small earthquakes (i.e., magnitude less than 4) are considered either point sources or homogeneous “penny‐shaped” surfaces. While these assumptions may be valid, details of earthquake ruptures are more complex. Here we study a magnitude 3 induced earthquake that occurred in the St. Gallen geothermal reservoir (NE Switzerland) in 2013. We image the earthquake rupture by refocusing on the fault the seismic energy recorded at six sensors located within 15 km from the source. Our results show that a detailed description of the rupture process of such a small earthquake can indeed be obtained: The rupture propagates from the hypocenter in NNE direction for 150 m, with an average velocity of 2 km/s, breaking into a less active portion of the fault, where no earthquake was previously recorded. The proposed method could be routinely applied during geothermal reservoir operations to allow rapid assessment of fault structures involved in the reservoir creation process. Our approach reveals rupture complexity for a magnitude 3.3 fluid‐induced earthquake at 100‐m scaleThe rupture propagated toward NNE, breaking into a previously less active portion of the faultEarly aftershocks distribute around the main slip area, suggesting a stress transfer mechanism

In July 2013, a sequence of more than 340 earthquakes was induced by reservoir stimulations and well‐control procedures following a gas kick at a deep geothermal drilling project close to the city of St. Gallen, Switzerland. The sequence culminated in an ML3.5 earthquake, which was felt within 10–15 km from the epicenter. High‐quality earthquake locations and 3‐D reflection seismic data acquired in the St. Gallen project provide a unique data set, which allows high‐resolution studies of earthquake triggering related to the injection of fluids into macroscopic fault zones. In this study, we present a high‐precision earthquake catalog of the induced sequence. Absolute locations are constrained by a coupled hypocenter‐velocity inversion, and subsequent double‐difference relocations image the geometry of the ML3.5 rupture and resolve the spatiotemporal evolution of seismicity. A joint interpretation of earthquake and seismic data shows that the majority of the seismicity occurred in the pre‐Mesozoic basement, hundreds of meters below the borehole and the targeted Mesozoic sequence. We propose a hydraulic connectivity between the reactivated fault and the borehole, likely through faults mapped by seismic data. Despite the excellent quality of the seismic data, the association of seismicity with mapped faults remains ambiguous. In summary, our results document that the actual hydraulic properties of a fault system and hydraulic connections between its fault segments are complex and may not be predictable upfront. Incomplete knowledge of fault structures and stress heterogeneities within highly complex fault systems additionally challenge the degree of predictability of induced seismicity related to underground fluid injections. A pre‐Mesozoic fault segment was reactivated during reservoir stimulation and gas kickHydraulic connectivity exists between borehole and fault across different lithologiesResults document complexities in structure, hydraulic condition, and stress of targeted fault zones

The failures of two recent deep geothermal energy projects in Switzerland (Basel, 2006; St. Gallen, 2013) have again highlighted that one of the key challenges for the successful development and operation of deep underground heat exchangers is to control the risk of inducing potentially hazardous seismic events. In St. Gallen, after an injection test and two acid injections that were accompanied by a small number of micro‐earthquakes (ML<0.2), operators were surprised by an uncontrolled gas release from the formation (gas kick). The “killing” procedures that had to be initiated following standard drilling procedures led to a ML3.5 earthquake. With ambient seismic noise cross correlations from nine stations, we observe a significant loss of waveform coherence that we can horizontally and vertically constrain to the injection location of the fluid. The loss of waveform coherence starts with the onset of the fluid injections 4 days prior to the gas kick. We interpret the loss of coherence as a local perturbation of the medium. We show how ambient seismic noise analysis can be used to assess the aseismic response of the subsurface to geomechanical well operations and how this method could have helped to recognize the unexpected reservoir dynamics at an earlier stage than the microseismic response alone, allowed. Using ambient noise for reservoir monitoringWaveform decoherence related to well operationsObservation of aseismic medium perturbations

Secondary triggering of aftershocks is widely observed and often ascribed to secondary static stress transfer. However, small to moderate earthquakes are generally disregarded in estimates of Coulomb stress changes (ΔCFS), either because of source parameter uncertainties or a perceived lack of importance. We use recently published high‐quality focal mechanisms and hypocenters to reassess the role of small to moderate earthquakes for static stress triggering of aftershocks during the 1992 Mw7.3 Landers, California, earthquake sequence. We compare the ΔCFS imparted by aftershocks (2≤M≤6) onto subsequent aftershocks with the total ΔCFS induced by the M>6 main shocks. We find that incremental stress changes between aftershock pairs are potentially more often positive than expected over intermediate distances. Cumulative aftershock stress changes are not reliable for receivers with nearby aftershock stress sources because we exclude unrealistic aftershock stress shadows that result from uniform slip models. Nonetheless, 27% of aftershocks receive greater positive stress from aftershocks than from the main shocks. Overall, 85% of aftershocks are encouraged by the main shocks, while adding secondary stress encourages only 79%. We infer that source parameter uncertainties of small aftershocks remain too large to convincingly demonstrate (or rule out) that secondary stress transfer induces aftershocks. An important exception concerns aftershocks in main shock stress shadows: well‐resolved secondary stress from detected aftershocks rarely compensates negative main shock stress; these aftershocks require a different triggering mechanism. Weak but significant evidence for secondary static stress triggering27% of aftershocks receive more positive secondary stress than from main shocksSecondary stress cannot explain most aftershocks in main shock stress shadows

—The test that Kagan and Jackson (1991, 1995) applied to the seismic gap hypothesis did not bring us closer to understanding the generation of large earthquakes. On the contrary, it led some to the conclusion that the rebound theory of earthquake generation should be rejected. We disagree with this point of view and argue that a global test of the simplified gap hypothesis cannot be done because it cannot account for differences in the slip history of fault segments and tectonic differences between separate plate boundaries. Kagan and Jackson did show, however, that the original gap hypothesis was oversimplified and should be refined. We propose that consideration of all the facts, including slip history and seismicity patterns in the Andreanof Islands, show that the concept of seismic gaps and the elastic rebound theory are correct for that segment of the plate boundary. The coseismic slip in the Mw8.7 earthquake that broke this plate boundary segment in 1957 was only 2 m, as published before the repeat earthquake of 1986 (Mw8), and thus, using a plate convergence rate of 7.3 cm/year, the return time in this cycle was expected to be less than 30 years, unless substantial aseismic creep occurs. This supports the time predictable model of mainshock recurrence. In addition, Kisslinger et al.(1985) and Kisslinger (1986) noticed a seismic quiescence in the subsequent source volume before the 1986 earthquake and attempted to predict it. The specific parameters he estimated were not entirely correct although his interpretation of the observed quiescence as a precursor was. We conclude that the 1986, Mw8, Andreanof earthquake was not an example that disproves the seismic gap hypothesis. On the contrary, it shows that the hypothesis that plate motions reload plate boundaries after most of the elastic energy is released in great ruptures was correct in this case. This suggests that great earthquakes occur preferably in mature gaps. We believe the testing of the seismic gap hypothesis by algorithm on a global scale is an example that illustrates that overly simplified tests can lead to erroneous conclusions. To make progress in the actual understanding of the physics of the process of great earthquake ruptures, one must consider all the facts known for case histories.

—By analyzing data from two catalogs of seismicity, one local and one regional, examples are drawn that demonstrate artificial seismicity patterns which can arise when changes in magnitude reporting or in detection ability are introduced by changes in operating procedures of a seismic network. In the first case, a catalog which comprises seismic data from Guerrero, Mexico, is revised and a correction is applied to magnitudes suspected of being affected by different operative practices. Next, several time windows are analyzed which compare the seismicity rate within the window to that of the total record mapped as a function of space. The same procedure is applied to the original (i.e., uncorrected) catalog. A significant seismic quiescence apparent in the original data set in the center of the seismic network all but disappears in the corrected data, indicating that this anomaly is not naturally induced. In the second case, we investigate the homogeneity of seismicity reporting for the Interior of Alaska. We computed the standard deviate zas a function of time by comparing the overall seismicity rate with the rate in a 3-year window. The maps z-values were inspected for all times. The most outstanding rate change is found around 1992.5. Since the b-value remained unchanged, the most reasonable explanation for the observed rate change around mid-1992 is a decrease in the detection ability of the network in the Interior of Alaska. Both case studies demonstrate the usefulness of systematic comparisons of the cumulative and noncumulative frequency-magnitude distribution, and of spatial and temporal mapping of the seismicity rates as a tool to investigate the homogeneity of earthquake reporting.

A systematic decay of the aftershock rate over time is one of the most fundamental empirical laws in Earth science. However, the equally fundamental effect of a mainshock on the size distribution of subsequent earthquakes has still not been quantified today and is therefore not used in earthquake hazard assessment. We apply a stacking approach to well‐recorded earthquake sequences to extract this effect. Immediately after a mainshock, the mean size distribution of events, or bvalue, increases by 20–30%, considerably decreasing the chance of subsequent larger events. This increase is strongest in the immediate vicinity of the mainshock, decreasing rapidly with distance but only gradually over time. We present a model that explains these observations as a consequence of the stress changes in the surrounding area caused by the mainshocks slip. Our results have substantial implications for how seismic risk during earthquake sequences is assessed. The effect of a mainshock on the size distribution of subsequent earthquakes has not been quantified and is therefore not used in earthquake hazard assessment. To quantify this effect, we develop a stacking approach centered on the mainshock time and apply it to for 31 well‐recorded aftershock sequences from around the world. We found that after a mainshock the earthquake size distribution shifts toward relative more smaller events, increasing the so‐called bvalue by 20–30%. One of the consequences of our finding is that the rates of large aftershocks are overestimated by the currently used models. Our result is fully consistent with both laboratory measurements and modeling, and we present a conceptual model that explains our findings. We develop a stacking approach to bvalue time series centered on the mainshock time in order to extract the generic behaviorApplying this approach to well‐recorded aftershock sequences, we demonstrate that the bvalue increases by 20–30% after a mainshockWe develop a Coulomb stress‐based model explaining the postmainshock bvalue increase and propose an empirical relationship to be used to forecast aftershock hazard

In this study, we present a Bayesian hierarchical framework to model fluid‐induced seismicity. The framework is based on a nonhomogeneous Poisson process with a fluid‐induced seismicity rate proportional to the rate of injected fluid. The fluid‐induced seismicity rate model depends upon a set of physically meaningful parameters and has been validated for six fluid‐induced case studies. In line with the vision of hierarchical Bayesian modeling, the rate parameters are considered as random variables. We develop both the Bayesian inference and updating rules, which are used to develop a probabilistic forecasting model. We tested the Basel 2006 fluid‐induced seismic case study to prove that the hierarchical Bayesian model offers a suitable framework to coherently encode both epistemic uncertainty and aleatory variability. Moreover, it provides a robust and consistent short‐term seismic forecasting model suitable for online risk quantification and mitigation. In this study, the authors propose a general and novel probabilistic framework for analyzing and forecasting uncertainties related to fluid‐induced seismicity. Their findings show that the proposed framework enables a novel and more in‐depth understanding of the uncertainties governing fluid‐induced seismicity. Moreover, they show that a new short‐time forecast model formulated using the proposed framework supports an accurate prediction of the number and magnitude of fluid‐induced events. Bayesian framework developed for classifying, analyzing, and forecasting uncertainties related to fluid‐induced seismicityRobust classification and treatment of epistemic and aleatory uncertainties in a nonhomogeneous Poisson frameworkShort‐term forecast model accurately predicts the number and maximum magnitude of events

We combine two complementary seismological methods to study the deep roots of arc volcanism in the Alaskan subduction zone: targeted seismic body wave tomography and imaging of the relative size distribution (bvalue) of earthquakes. For the tomography we apply a staggered inversion scheme, starting with the minimum 1‐D velocity model, progressing to a 3‐D velocity model. Inversions are based on traveltime data from about 5500 events (1971–2002) with approximately 100,000 Pand 50,000 Shigh‐quality arrivals, allowing us to resolve the 3‐D velocity field on a 20 × 20 km grid. For the bvalue imaging we have introduced a new 3‐D sampling approach to map the bvalues on surfaces in the Wadati‐Benioff zone (WBZ). Our bvalue study is based on 12,474 relocated events of magnitude of completeness ≥2.4 and depth >40 km. We observe anomalously high bvalues at depths around 100 km, appearing locally concentrated beneath active subduction volcanoes. We believe that these zones are associated with dehydration of the slab and subsequent magma generation. The analysis of VP/VSratios likewise shows indications of the presence of fluids: significantly higher VP/VSratios rising in columns from the top of the WBZ at 100 km depth below active volcanic centers. On the basis of the joint interpretation, we propose that our images track fluids from their genesis at 100 km depth at the top of the subducting plate to the bottom of the crust below volcanoes.